Satellite Image Classification with Fuzzy Logic: from Hard to Soft Computing Situation

Transcription

1 Volume 1, No. 9, November 2012 ISSN The International Journal of Computer Science & Applications (TIJCSA) RESEARCH PAPER Available Online at Satellite Image Classification with Fuzzy Logic: from Hard to Soft Computing Situation A. L. Choodarathnakara Dept. of Electronics & Communication Engineering Government Engineering College Kushalnagar , INDIA Dr. T. Ashok Kumar 2, Dr. Shivaprakash Koliwad 3, Dr. C. G. Patil 4 2, 3 Dept. of Electronics & Communication Engineering and 4 Master Control Facility (MCF) 2 Vivekananda College of Engg. & Technology, Puttur (DK), 3 MCE and 4 MCF, Hassan, INDIA 2 ashokkumar1968@rediffmail.com, 3 spksagar2006@yahoo.co.in, 4 dir@mcf.gov.in Abstract Fuzzy logic, a knowledge-based method, widely used in Pattern Recognition today and is proposed to be applied in remote sensing image classification. Fuzzy logic makes no assumption about statistical distribution of the data and it provides more complete information for a thorough image analysis, such as fuzzy classification results. It is interpretable and can use expert knowledge and training data at the same time. Major advantage of this theory is that it allows the natural description in linguistic terms of problems that should be solved rather than in terms of relationships between precise numerical values. This advantage, dealing with the complicated systems in simple way, is the main reason why fuzzy logic theory is widely applied in decision making. Also, it is possible to classify the remotely sensed image as well as any other digital imagery; in such a way that certain land cover classes are clearly represented in the resulting image. The urban land cover types show spectral similarity in remote sensing data. The finer the spatial resolution of the data, the larger is the number of detectable subclasses within classes. This high withinclass spectral variance of some classes results in multimodal distribution of spectra and may decrease their spectral separability. Hence, the existing traditional hard classification techniques which are parametric type do not perform well on high resolution data in the complex environment of the urban area as they expect datasets to be distributed normally. Hence this paper is to investigate fuzzy soft classifier as an alternative to traditional hard classifier to classify satellite image data of a semi urban area. Keywords- Image Classification, Soft Computing, Supervised Classification, Unsupervised Classification, Artificial Neural Networks, Decision Tree, Fuzzy Logic, Accuracy Assessment. 2012, - TIJCSA All Rights Reserved 101

2 1. Introduction Remote Sensing (RS) refers to the science of identification of earth surface features and estimation of their geo-biophysical properties using electromagnetic radiation as a medium of interaction. Spectral, spatial, temporal and polarization signatures are major characteristics of the sensor/target, which facilitate target discrimination. Earth surface data as seen by the sensors in different wavelengths (reflected, scattered and/or emitted) is corrected radiometrically and geometrically before extraction of spectral information. RS data, with its ability for a synoptic view, repetitive coverage with calibrated sensors to detect changes, observations at different resolutions, provides a better alternative for natural resources management as compared to traditional methods. Some of the major operational application themes, in which India has extensively used remote sensing data, are agriculture, forestry, water resources, land use, urban sprawl, geology, environment, coastal zone, marine resources, snow and glacier, disaster monitoring and mitigation, infrastructure development, etc [1], [9]. Image classification is the process of categorizing all the pixels automatically in an image into a finite number of land cover classes [12] and it is one of the most often used quantitative data analysis techniques in remote sensing to describe ground cover types or material classes. Classifiers are broadly categorized into supervised and unsupervised, hard and soft, parametric and non parametric type. Among them, the maximum likelihood classifier (MLC), belonging to the family of supervised parametric classifier is most commonly used in remote sensing because of its robustness and easy availability in almost all image processing software [1]. Also, MLC has traditionally been employed as a baseline for evaluating the accuracy of classifiers on remotely sensed data [10], [15], [16]. For classification of features in urban area, the expected spatial resolution should be at least 5m where buildings and roads can be easily distinguished [3]. Most of the materials found in the urban environment like concrete, asphalt, metal, plastic, glass, water etc. exhibit spectral similarity. Many urban land cover types such as roads, buildings, parking lots, grass, trees, shrubs and soil also show spectral similarity. In addition to the spectral similarity between land cover types, remote sensing images contain mixed pixels which make it difficult to classify a pixel as belonging to only one class [4], [6]. Therefore, the finer the spatial resolution, the larger is the number of detectable subclasses within the classes and this high within-class spectral variance of some classes may decrease their spectral separability resulting in lower classification accuracy. As a result, classifications accuracies may decrease for some classes, such as complex urban are as spatial resolution becomes finer [5]. As the existing traditional hard classification techniques are parametric type, they do not perform well on high resolution data in the complex environment of the urban area as they expect datasets to be distributed normally [2], [4]. The assumption of normal distribution of spectra is often violated especially in the complex landscapes in high-resolution data. In addition, insufficient, non-representative, or multimode distributed training samples can further introduce uncertainty to the image classification procedure. Another major drawback of the parametric classifiers lies in the difficulty of integrating spectral data with 2012, - TIJCSA All Rights Reserved 102

3 ancillary data [1], [2] like digital elevation model, slope, texture and context information, etc. Therefore, parametric (hard) classifiers fail to exploit the best use of the information available through advanced sensor systems and various ancillary data. Hence, generating a satisfactory classified image from a high-resolution remotely sensed data is a challenge and is not as straightforward as classification of low resolution imagery (30 m or more) using traditional classifiers. On the contrary, non-parametric classifiers are independent of the properties of the distribution of data. With nonparametric classifiers, the assumption of the normal distribution of the dataset is not required. No statistical parameters are needed to separate image classes. Nonparametric classifiers are thus suitable for the incorporation of non-spectral data into a classification procedure. Among the most commonly used nonparametric classification approaches are Fuzzy Logic, Neural Networks, Decision Trees, Support Vector Machines, and Expert Systems [1]. 2. Image Classification Image classification is a complex process that may be affected by many factors. Effective use of multiple features of remotely sensed data and the selection of a suitable classification method are especially significant for improving classification accuracy. Non-parametric classifiers such as fuzzy logic, neural network, decision tree classifier, and knowledge-based classification have increasingly become important approaches for multisource data classification. Integration of remote sensing, geographical information systems (GIS), and expert system emerges as a new research frontier. More research, however, is needed to identify and reduce uncertainties in the image-processing chain to improve classification accuracy [10], [15]. In general, image classification approaches can be grouped as supervised and unsupervised, or parametric and nonparametric, or hard and soft (fuzzy) classification, or per-pixel, sub-pixel, and per-field. Based on whether training samples used or not used image classification can be of two types: A. Supervised Classification In this type of classification the image analyst supervises the pixel categorization process by specifying, to the computer algorithm, numerical descriptors of the various land cover types present in a scene. To do this representative sample sites of known cover type, called training areas, are used to compile a numerical interpretation key that describes the spectral attributes for feature type of interest. Each pixel in the data set is then compared numerically to each category in the interpretation key and labelled. There are a number of numerical strategies that can be employed to make this comparison between unknown pixels and training set pixels. Examples of supervised classification Maximum likelihood, minimum distance, artificial neural network, decision tree classifier [15]. 2012, - TIJCSA All Rights Reserved 103

4 Figure1. Supervised Classification Scheme 1) Maximum Likelihood Classifier: The maximum likelihood classifier quantitatively evaluates both the variance and covariance of the category spectral response patterns when classifying an unknown pixel. An assumption that the distribution of the cloud of points forming the category training data is Gaussian is made. With this assumption the distribution of a category response pattern can be completely described by the mean vector and the covariance matrix. Given these parameters statistical probability of a given pixel value being a member of a particular land cover class can be computed [15], [16]. In essence, the maximum likelihood classifier delineates ellipsoidal equi-probability contours in the scatter diagram and these decision regions are shown in Figure 2. The shape of the equi-probability contours expresses the sensitivity of the likelihood classifier to covariance. For example because of the sensitivity it can be seen that pixel 1 would be appropriately assigned to the corn category. Figure 2. Equi-probability contours defined by a maximum likelihood classifier 2) Minimum-Distance-To-Means-Classifier: One of the simpler classification strategies is shown in Figure 3 called Min-Distance-To- Mean classifier. First, the mean, or average, spectral value in each band for each category is determined. These values comprise the mean vector for each category. The category means is indicated by + s symbol as shown in Figure 3. By considering the two channel pixel values as positional coordinates, a pixel of unknown identity may be classified by computing the distance between the value of the unknown pixel and each of the category means. In Figure 3, an unknown pixel has been plotted at a point 1. The distance between 2012, - TIJCSA All Rights Reserved 104

5 this pixel value and each category is illustrated by dashed lines. In this case the class is corn. The pixel value plotted at point 2 is identified as sand. Figure 3. Minimum distance to means classification strategy B. Unsupervised Classification Unsupervised classification (commonly referred to as clustering) is an effective method of partitioning remote sensor image data in multispectral feature space and extracting landcover information. Compared to supervised classification, unsupervised classification normally requires only a minimal amount of initial input from the analyst. This is because clustering does not normally require training data. The unsupervised procedures are applied in two separate steps. In the unsupervised approach the image data are first classified by aggregating them into the natural spectral groupings or clusters as shown in Figure 4. Then the image analyst determines the land cover identity of these spectral groups by comparing the classified image data to ground reference data [10], [23]. Figure 4. Unsupervised Classification Scheme The following are the examples of unsupervised classifiers: 1) K-means Clustering Algorithm: An initial mean vector (seed) is arbitrarily specified for each of K clusters; each pixel is then assigned to the class whose mean vector is closest to the pixel vector; A new set of cluster mean vector is then calculated from this classification, and pixels are reassigned accordingly. In each iteration, the K-means will tend to migrate to the true concentrations of data. The iterations are continued until there is no significant change in terms of the net mean migration from one iteration to the next. The final, stable result is not sensitive to the initial specification of seed vector, but more iteration may be required for convergence if 2012, - TIJCSA All Rights Reserved 105

6 the final vectors are not close to the seed vectors. The final cluster mean vectors may be used to classify the entire image with a minimum-distance classifier in one additional pass, or the covariance matrices of the clusters may be calculated and used with the mean vectors in a maximum-likelihood classification. 2) ISODATA Clustering: The Iterative Self-Organizing Data Analysis Technique (ISODATA) represents a comprehensive set of heuristic (rule of thumb) procedures that have been incorporated into an iterative classification algorithm. Many of the steps incorporated into the algorithm are a result of experience gained through experimentation. ISODATA is self-organizing because it requires relatively little human input. The ISODATA algorithm is a modification of the k-means clustering algorithm. It includes merging clusters if their separation distance in multi-spectral feature space is below a user-specified threshold and Rules for splitting a single cluster into two clusters. ISODATA is iterative because it makes a large number of passes through the remote sensing dataset until specified results are obtained, instead of just two passes. Better results will be obtained if all bands have the similar data ranges. C. Hybrid Classification This technique involves the aspects of both supervised and unsupervised classification and are aimed at improving the accuracy or efficiency (or both) of the classification process. Hybrid classifiers are valuable in analyses where there is complex variability in the spectral response patterns for individual cover types present Guided clustering is a hybrid approach which is used in applications like vegetation mapping. In guided clustering the analyst delineates numerous supervised-like training sets for each cover type to be classified. The data from all the training sites are then used in an unsupervised clustering routine to generate several spectral signatures. These signatures are examined by the analyst; some may be discarded or merged and the remainder is considered to represent spectral subclasses of the desired information class. Once sufficient subclasses are obtained, a maximum likelihood classification is performed. The spectral classes are then aggregated back into the information classes. D. Per-pixel classification approaches Traditional per-pixel classifiers typically develop a signature by combining the spectra of all training-set pixels for a given feature. The resulting signature contains the contributions of all materials present in the training pixels, but ignores the impact of the mixed pixels. Per-pixel classification algorithms can be parametric or non-parametric. The parametric classifiers assume that a normally distributed dataset exists, and that the statistical parameters (e.g. mean vector and covariance matrix) generated from the training samples are representative. However, the assumption of normal spectral distribution is often violated, especially in complex landscapes. In addition, insufficient, non-representative, or multimode distributed training samples can further introduce uncertainty to the image classification procedure. Another major drawback of the parametric classifiers lies in the difficulty of integrating spectral data with ancillary data. The MLC may be the most commonly used parametric classifier in practice, because of its robustness and its easy availability in almost any image-processing software. 2012, - TIJCSA All Rights Reserved 106

7 E. Subpixel classification approaches Subpixel classification approaches have been developed to provide a more appropriate representation and accurate area estimation of land covers than per-pixel approaches, especially when coarse spatial resolution data are used. One major drawback of subpixel classification lies in the difficulty in assessing accuracy. F. Per-field classification approaches The heterogeneity in complex landscapes results in high spectral variation within the same land-cover class. With per-pixel classifiers, each pixel is individually grouped into a certain category, and the results may be noisy due to high spatial frequency in the landscape. The per-field classifier is designed to deal with the problem of environmental heterogeneity, and has shown to be effective for improving classification accuracy. The per-field classifier averages out the noise by using land parcels (called fields ) as individual units Geographical information systems (GIS) provide a means for implementing per-field classification through integration of vector and raster data. The vector data are used to subdivide an image into parcels, and classification is then conducted based on the parcels, thus avoiding intra-class spectral variations. 3. Fuzzy Logic in Image Classification A fuzzy set is a set whose elements have degrees of membership. A element of a fuzzy set can be full member (100% membership) or a partial member (between 0% and 100% membership). That is, the membership value assigned to an element is no longer restricted to just two values, but can be 0, 1 or any value in-between. Mathematical function which defines the degree of an element's membership in a fuzzy set is called membership function. The natural description of problems, in linguistic terms, rather than in terms of relationships between precise numerical values is the major advantage of this theory. Let X be a universe of discourse, whose generic elements are denoted x. Thus, X = {x}. Membership in a classical set A of X is often viewed as a characteristic function x A from X to {0, 1} such that x A (x) = 1 if and only if x belongs to A. A fizzy set B in X is characterized by a membership function j which associates with each x a real number in [0, 1]. f B represents the grade of membership of x in B. The closer the value of f B (x) is to 1, the more x belongs to B. Fuzzy set theory can provide a better representation for geographical information, much of which cannot be described well by a single class. In a fuzzy representation for remote sensing image analysis, land-cover classes can be defined as fuzzy sets, and pixels as elements of set. Each pixel is attached with a group of membership grades to indicate the extent to which the pixel belongs to certain classes. Pixels with class mixture or in intermediate conditions can be described by membership grades. For example, if a ground cell contains two cover-types, soil and vegetation, it may have two membership grades indicating the extents to which it is associated with the two classes [18], [19], [22], [23]. A. Fuzzy Representation of Geographical Information: 2012, - TIJCSA All Rights Reserved 107

8 Geographical information is conventionally represented in thematic maps. A map is a set of points, lines and areas that are defined both by their location in space and by their nonspatial attributes. The primary objective of many remote sensing applications is the classification of images to prepare thematic maps and the information required for the classification is derived from thematic maps. Each pixel can only be associated with a single cover class. Such a method cannot properly represent class mixture and intermediate conditions which occur in almost remote sensing images. A pixel corresponds to a cell on a ground. Quite often such a cell contains mixture of surface cover classes, for example, grass and underlying soil. Since currently one cover class can be applied to pixel information about other component classes and deviation of the assignment cannot be represented. Different conditions may exist within a cover class. For example, vegetation may be in different conditions that are caused by such factors as plant health, age, and water content. However, these conditions currently cannot be differentiated unless more classes are defined. It is clearly inaccurate to assign the same class to fresh grass and half dry grass without specifying their differences. Introducing more classes will lead to higher analysis costs and, no matter how finely the classes are defined, within-class variability may exist. Fuzzy sets, which constitute the oldest component of soft computing, are suitable for handling the issues related to understandability of patterns, incomplete/noisy data, mixed media information and human interaction and can provide approximate solutions faster. The development of fuzzy logic has lead to the emergence of soft computing. A popular approach to derive soft classifiers is Fuzzy Logic (FL). FL extends the conventional Boolean Logic using the concept of partial truth for those values falling between 0 and 1, which corresponds to a totally true and totally false, respectively. This method consists of mathematical tools, to model approximate reasoning when data is imprecise, uncertain, vague, and incomplete, using fuzzy sets. In Crisp Logic, an element in the universe is defined as a member or not of a given set (Tsoukalas 1997). For instance, the membership of a crisp set H can be defined through a membership function defined for every element y of the universe as (1) On the other hand, in fuzzy sets members have degree of membership in the set. The membership in a set is represented by a number in the range from 0 to 1. (2) Where, y is an element of the universe. They have been mainly used in clustering, discovering association rules and functional dependencies, data summarization, time series analysis, web applications and image retrieval. The modelling of imprecise and qualitative knowledge, as well as the transmission and handling of uncertainty at various stage sis possible through the use of fuzzy sets. B. Fuzzy Clustering Data mining aims at sifting through large volumes of data in order to represent useful information in the form of new relationships, patterns, clusters, for decision making by a 2012, - TIJCSA All Rights Reserved 108

9 user. Fuzzy sets support a focused search, specified in linguistic terms, through data. They also help discover dependencies between the data in qualitative/semi qualitative format. In data mining, discovery of structure and an eventual quantification of functional dependencies as led to the development of fuzzy clustering algorithms. Achieving accuracy is focus of important in data mining because there are too many attributes and values to be considered and can result in combinatoric explosion. A soft focus is used to handle both crisp and imprecise data. Non-crisp values are handled by granularization followed by partitioning. Increased granularity reduces attribute distinctiveness, resulting in loss of useful information, while finer grains lead to partitioning difficulty. Soft granules can be defined in terms of membership functions. Granular computing is useful in finding meaningful patterns in data by expressing and processing chunks of information (granules). 4. Fuzzy Soft Classification Scheme Fuzzy logic system is an automatic system that is capable of mimicking human actions for a specific task. There are three main operations in a fuzzy logic system. The first operation is fuzzification, which is the mapping from a crisp point to a fuzzy set. The second operation is inferencing, which is the evaluation of the fuzzy rules in the form of IF-THEN. The last operation is defuzzification, which maps the fuzzy output of the expert system into a crisp value, as shown in Figure 5. The figure 5 also represents a definition of a fuzzy expert system [16]. Figure 5. Fuzzy Logic Expert System Fuzzy systems are being used successfully in an increasing number of application areas. They use linguistic rules to describe systems. The rule based systems are more suitable for complex systems where it is very difficult, but not impossible to describe the system mathematically. The basic structure of a fuzzy technique includes four main components. A fuzzifier, which converts crisp values (real time values) into fuzzy values An inference engine, that applies a fuzzy reasoning mechanism to obtain a fuzzy output A defuzzifier, which translates this later output into crisp values A knowledge base which contains both an ensemble of fuzzy rules known as rule base and an ensemble of membership functions known as database. 2012, - TIJCSA All Rights Reserved 109

10 Unlike traditional Aristotelian two valued logic, in fuzzy logic, fuzzy set membership occurs by degree over the range [0, 1], which is represented by a membership function. This is the membership function of that fuzzy set. A. Fuzzy Inference System Fuzzy inference is the process of formulating the mapping from a given input to an output using fuzzy logic. The process of fuzzy inference involves: membership functions, fuzzy logic operators and if-then rules. 1) Membership Function Membership function is the mathematical function which defines the degree of an element's membership in a fuzzy set. 2) Fuzzy Logic operators The most important thing to realize about fuzzy logical reasoning is the fact that it is a superset of standard Boolean logic. In other words, if the fuzzy values are kept at their extremes of 1 (completely true) and 0 (completely false), standard logical operations will hold. That is, A AND M operator is replaced with minimum - min (A, M) operator, A OR M with maximum - max (A, M) and NOT M with 1-M. 3) If-Then rules Fuzzy sets and fuzzy operators are the subjects and verbs of fuzzy logic. Usually the knowledge involved in fuzzy reasoning is expressed as rules in the form: If x is A Then y is B (3) Where, x and y are fuzzy variables and A and B are fuzzy values. The if-part of the rule "x is A" is called the antecedent or premise, while the then-part of the rule "y is B" is called the consequent or conclusion. Statements in the antecedent (or consequent) parts of the rules may well involve fuzzy logical connectives such as AND and OR. In the if-then rule, the word "is" gets used in two entirely different ways depending on whether it appears in the antecedent or the consequent part. Based on whether the output definitive decision about land cover class or not the classification is of two types: B. Hard Classification In hard (crisp) classification each pixel is forced or constrained to show membership to a single class as shown in Figure 6. The characteristics of Hard classification are: Making a definitive decision about the land cover class that each pixel is allocated to a single class. The area estimation by hard classification may produce large errors, especially from coarse spatial resolution data due to the mixed pixel problem. Examples of hard classifiers are maximum likelihood, minimum distance, artificial neural network, decision tree, and support vector machine. 2012, - TIJCSA All Rights Reserved 110

11 Figure 6. Example of Hard Classification C. Fuzzy (Soft)Classification In soft (fuzzy) classification each pixel may display multiple and partial class membership as shown in figure 7. Soft classification has been proposed in the literature as an alternative to hard classification because of its ability to deal with mixed pixels. The characteristics of soft classification are: Providing for each pixel a measure of the degree of similarity for every class. Soft classification provides more information and potentially a more accurate result, especially for coarse spatial resolution data classification. Examples of soft classifiers are: Fuzzy-set classifiers, subpixel classifier, and spectral mixture analysis. Figure 7. Example of Soft Classification D. Accuracy Assessment 2012, - TIJCSA All Rights Reserved 111

12 Classification process is not complete until its accuracy is assessed. Accuracy assessment can be performed by comparing two sources of information (Jensen, 1996): Remotesensing derived classification data and reference test data. The relationship of these two sets is summarized in an error matrix where columns represent the reference data while rows represent the classified data. An error matrix is a square array of numbers laid out in rows and columns that expresses the number of sample units assigns to a particular category relative to the actual category as verified in the field. The accuracy of a classification has traditionally been measured by the overall accuracy. The overall accuracy of classification is obtained by dividing the sum of the correctly classified pixels (i.e. summed up values on the major diagonal of the error matrix) by the total number of pixels classified of the reference points. The kappa statistic, also called KHAT value, is a measure of how well the classification agrees with the reference data. It is also a measure of overall accuracy [13] and most commonly employed to evaluate the performance of a classifier. But the overall accuracy alone gives no insight into how well the classifier is performing for each of the different classes. In particular, a classifier might perform well for a class which accounts for a large proportion of the test data and this will bias the overall accuracy, despite low class accuracies for other classes. To avoid such a bias, it is important to consider the individual class accuracy under producer s accuracy and user s accuracy. Producer s accuracy is a measure of the probability of a reference pixel being correctly classified and also called a measure of omission error. It is obtained by dividing the total number of correct pixels in a category by the total number of pixels of that category as derived from the reference data [14]. User s accuracy can be obtained by dividing the total number of correct pixels in a category by the total number of pixels that were classified in that category and also called a measure of commission error. It is an indicative of the probability that a pixel classified on the image actually represents that category on the ground [21]. 5. Conclusion The fuzzy technique will improve the analysis accuracy considerably and also more information about land cover and land change can be obtained easily. Although many classification approaches have been developed, which approach is suitable for features of interest in a given study area is not fully understood. Classification algorithms can be perpixel, sub-pixel, and per-field. Per-pixel classification is still most commonly used in practice. However, the accuracy may not meet the requirement of research because of the impact of the mixed pixel problem. Sub-pixel algorithms have the potential to deal with the mixed pixel problem, and may achieve higher accuracy for medium and coarse spatial resolution images. Per-field classification approaches are most suitable for fine spatial resolution data. When using multisource data, such as a combination of spectral signatures, texture and context information, and ancillary data, advanced non-parametric classifiers such as fuzzy logic, neural network, decision tree, and knowledge-based 2012, - TIJCSA All Rights Reserved 112

13 classification, may be more suited to handle these complex data processes, and thus have gained increasing attention in the remote-sensing community in recent years. Acknowledgement The author wish to acknowledge the people who were helped me in preparing this paper. Also, the author wants to acknowledge the guide Dr. Shivapraksh Koliwad for his novel suggestions and author thanks teaching, non-teaching and supporting staff of Malnad College of Engineering, Hassan and Government Engineering College, Kushalnagar. References [1] Navalgund, R. R., Remote sensing: Basics and applications, Resonance, 2001, 6, [2] Patel N. K., Medhavy T. T., Patnaik C. and Hussain A., Multi-temporal ERS-1 SAR data for identification of rice crop, J. of Indian Soc. R. S., 1995, 23, [3] Panigrahy S., Chakraborty M., Sharma S. A., Kundu N., Ghose S. C. and Pal M., Early estimation of rice acre using temporal ERS-1 synthetic aperture radar data A case study for Howrah and Hooghly districts of West Bengal, India, Int. J. Remote Sensing, 1997, 18, [4] Srivastava H. S., Patel P. and Navalgund R. R., Incorporating soil texture in soil moisture estimation from extended low-1 beammode Radarsat-SAR data, Int. J. Remote Sensing, 2006, 27, [5] Jayaraman V., Srivastava S. K., Kumaran Raju K. and Rao U.R., Total solution approach using IRS-1C and IRS-P3: A perspective of multi-resolution data fusion and improved vegetation indices, IEEE Trans. Geosci. R. S., 2000, 38, [6] Navalgund R. R., Parihar J. S., Ajai and Nageshwar Rao P. P., Crop inventory using remotely sensed data, Curr. Sci., 1991, 61, [7] Vasudevan B. G., Gohil B. S. and Agarwal V. K., Backpropagation neural network based retrieval of atmospheric water vapour and cloud liquid water from IRS-P4 MSMR, IEEE Trans Geosci. Remote Sensing, 2004, 42(5), [8] C. Apte and S. Weiss, Data mining with decision trees and decision rules, Future Generation Computer Systems, pp , Nov [9] Selim Aksoy, K. Koperski, Carsten Tusk, and Giovanni Marchisio, Interactive training of advanced classifiers for mining remote sensing image archives, KDD 04, Seattle, Washington, USA, August 22-25,2004. [10] Jiawei Han and Micheline Kamber, Data Mining Concepts and Techniques, 2nd ed., Elsevier, [11] S. Rasoul Safavian and David Landgrebe, A survey of decision tree classifier methodology, IEEE Trans. Syst. Man Cybnet., vol. 21, no. 3, pp , May/ June [12] Lior Rokach and Oded Maimon, Top-down induction of decision trees classifiers-a survey, IEEE Trans. Syst. Man Cybnet., Part-C, Vol. 35, No. 4, pp , Nov [13] R. G. Congalton and A. Roy Mead, A review of three discrete multivariate analysis techniques used in assessing the accuracy of remotely sensed data from error matrix, IEEE Trans. Geosci. Remote Sensing, Vol. GE 24, No. 1, pp , Jan , - TIJCSA All Rights Reserved 113

14 [14] R. G. Congalton, Considerations and techniques for assessing the accuracy of remotely sensed data, Geosci. Remote Sens. Symp., Vol. 3, Jul , 1989, pp [15] Barandela R. and M. Juraez, Supervised Classification of Remotely Sensed Data with Ongoing Learning Capability, IJRS,, Vol. 23, No. 22, 2002, pp [16] Foody G. M., 2002a, Hard and soft classifications by a neural network with a nonexhaustively defined set of classes, International Journal of Remote Sensing, 23, pp [17] P. D. Heermann and N. Khazenic, Classification of Multispectral Remote Sensing Data using a Back- Propagation Neural Network, IEEE Trans. Geoscience and Remote Sensing, Vol. 30, No. 1, pp , Jan [18] F. Wang, Improving Remote Sensing Image Analysis through Fuzzy Information representation, Photogramm. Engg. R. S., Vol. 56, No. 8, pp ,1990. [19] G. M. Foody, A fuzzy sets approach to the representation of vegetation continua from remotely sensed data: An example from lowland heath, Photogramm. Engg. Remote Sens, Vol. 58, No. 2, pp , [20] Thomas M. Lillesand, Ralph W.Kiefer and Jonathan W. Chipman, Remote Sensing and Image Interpretation, Fifth Edition, 2007, pp [21] Congalton R. G., A review of assessing the accuracy of classification of remotely sensed data, Remote Sensing of Environment, 37, pp , [22] Foody G. M., Approaches for the production and evaluation of fuzzy land cover classification from remotely-sensed data, International Journal of Remote Sensing, 17, pp , [23] Franklin S. E. and Wulder M. A., Remote sensing methods in medium spatial resolution satellite data land cover classification of large areas, Progress in Physical Geography, 26, pp , [24] P. M. Atkinson and A. R. L. Tatnall, Neural networks in remote sensing: An Introduction, IJRS, vol. 18, , [25] Freeman, Artificial Neural Network Algorithm: Applications and Programming, Comp and Neural Systems Series, [26] F. Aminzadeh and M. Jamshidi, Soft Computing, Fuzzy Logic, Neural Networks and Distributed Artificial Intelligence, Prentice Hall, New Jersey, [27] L. A. Zadeh, Calculus of fuzzy restrictions, Fuzzy Sets and their Application to Cognitive and Decision Processes, IEEE Conference on Fuzzy Systems, , , - TIJCSA All Rights Reserved 114